Crouzeix's Conjecture in Matrix Analysis

• Crouzeix's Conjecture is a hypothesis in operator theory stating that the numerical range of a matrix bounds its holomorphic functional calculus with a sharp constant of 2.
• Recent advances, including the Crouzeix–Palencia bound, verify the conjecture for various matrix classes and reveal deep connections with spectral sets, conformal mappings, and extremal functions.
• Techniques such as Cauchy integrals, Blaschke products, and random matrix asymptotics provide practical insights for extending the conjecture and refining operator norm estimates.

Crouzeix’s conjecture is the assertion that the numerical range of a matrix controls its holomorphic functional calculus with the sharp universal constant $2$. For a square complex matrix $A$, with numerical range

$W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$

the conjecture states that

$\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$

for every polynomial $p$. It lies at the intersection of numerical range theory, spectral sets, operator model theory, and matrix function estimates. The best general theorem currently available is the Crouzeix–Palencia bound $1+\sqrt2$, while the sharp constant $2$ is known only in special classes and asymptotic regimes (Bickel et al., 2020).

1. Statement, variants, and basic framework

The conjecture is most commonly formulated for polynomials, but the same inequality is equivalent to corresponding statements for functions analytic on a neighborhood of $W(A)$, and also for functions in $\mathbf A(W(A))$, the algebra of functions analytic on $W(A)^\circ$ and continuous on $A$0. If it holds for matrices, then it extends to bounded operators on Hilbert space; if it holds for polynomials, then Mergelyan approximation extends it to suitable analytic functions on $A$1 (Bickel et al., 2020).

A standard spectral-set reformulation introduces, for an open set $A$2 containing $A$3,

$A$4

Crouzeix’s conjecture is precisely the claim that

$A$5

In this language, $A$6 is conjectured to be a $A$7-spectral set for every matrix. The strongest general theorem presently cited in the literature is

$A$8

the Crouzeix–Palencia bound (Chen et al., 2023).

A stronger, completely bounded version replaces scalar polynomials by matrix-valued polynomials $A$9 and asks for

$W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$0

Special-case proofs often establish this stronger form. In the same circle of ideas, an equivalent numerical-radius formulation is available: the conjectured constant $W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$1 is equivalent to the sharp bound

$W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$2

where $W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$3 is the numerical radius (Glader et al., 2017).

2. Spectral-set machinery and extremal structure

A central analytic tool is the Cauchy integral representation

$W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$4

valid when $W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$5 contains $W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$6. In the Crouzeix–Palencia framework one also defines a companion transform

$W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$7

and, after arclength parametrization $W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$8, the Hermitian kernel

$W(A)=\{\langle Ax,x\rangle:\|x\|=1\},$9

For $\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$0, positivity properties of $\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$1 yield the universal $\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$2 estimate. For regions strictly inside $\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$3, positivity may fail, and modified estimates introduce scalar correction terms built from the smallest eigenvalue of $\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$4 (Chen et al., 2023).

The geometry becomes especially transparent for disks. If $\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$5 contains $\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$6, then the associated scalar transform collapses to a constant,

$\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$7

and the Caldwell–Greenbaum–Li singular-vector argument yields the exact disk bound

$\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$8

This disk case is a recurring local model in perturbative arguments for “nearly circular” numerical ranges (Chen et al., 2023).

The extremal viewpoint is equally important. For a simply connected domain $\|p(A)\|\le 2\max_{z\in W(A)}|p(z)|$9, an extremal function for $p$0 is a function $p$1 maximizing $p$2. Such extremals have the form

$p$3

where $p$4 is conformal and $p$5 is a finite Blaschke product of degree at most $p$6. Associated extremal vectors satisfy strong orthogonality identities; for example, if $p$7, then an associated extremal unit vector $p$8 satisfies

$p$9

There is also a representation theorem: for an extremal pair $1+\sqrt2$0, there exists a unique probability measure $1+\sqrt2$1 on $1+\sqrt2$2 such that

$1+\sqrt2$3

with additional cancellation identities when $1+\sqrt2$4. These results clarify why extremal Blaschke products, conformal maps, and singular vectors are structurally central to the conjecture (Bickel et al., 2020).

3. Verified classes and sharp extremal mechanisms

A substantial literature verifies the conjectured constant $1+\sqrt2$5 for specific matrix classes. The following families are explicitly established in the cited works.

Matrix class	Hypothesis	Conclusion
$1+\sqrt2$6 matrices	none	conjecture holds
$1+\sqrt2$7 tridiagonal constant-diagonal matrices	equivalently, elliptic numerical range centered at an eigenvalue	complete $1+\sqrt2$8-spectral set
Contractions with separated eigenvalues	$1+\sqrt2$9	conjecture holds
Cyclic weighted shifts $2$0	none	complete $2$1-spectral set
Some unicritical compressed shifts	degree $2$2 or $2$3, explicit parameter ranges	conjecture holds

For contractions with distinct eigenvalues in $2$4, the relevant separation parameter is the pseudohyperbolic separation constant

$2$5

and if $2$6, then one obtains

$2$7

for every $2$8, hence the conjecture follows a fortiori since $2$9 (Bickel et al., 2020).

For $W(A)$0 matrices with elliptic numerical range centered at an eigenvalue, equivalently tridiagonal matrices with constant diagonal,

$W(A)$1

the conjecture is proved in the strong completely bounded form by combining an explicit conformal map from the ellipse to the disk with carefully constructed similarity transforms of condition number $W(A)$2 (Glader et al., 2017).

For cyclic weighted shifts

$W(A)$3

a recent alternative proof shows

$W(A)$4

so $W(A)$5 is a complete $W(A)$6-spectral set. Moreover, if $W(A)$7, then the constant is actually strict: $W(A)$8 The proof exploits rotational symmetry of $W(A)$9, the identity $\mathbf A(W(A))$0 for the normalized Riemann map $\mathbf A(W(A))$1, and an extremal-pair argument within the Crouzeix–Palencia decomposition (Crouzeix et al., 18 Aug 2025).

The sharpness problem is illuminated by the structure of half-radial matrices, defined by

$\mathbf A(W(A))$2

where $\mathbf A(W(A))$3 is the numerical radius. These are exactly the equality cases for the identity polynomial $\mathbf A(W(A))$4. They are characterized by a rigid block form

$\mathbf A(W(A))$5

with $\mathbf A(W(A))$6 and $\mathbf A(W(A))$7. For monomials $\mathbf A(W(A))$8, equality

$\mathbf A(W(A))$9

holds exactly when $W(A)^\circ$0 is half-radial and $W(A)^\circ$1. In dimension $W(A)^\circ$2, equality for $W(A)^\circ$3 characterizes the Crabb–Choi–Crouzeix matrix, and more generally monomial extremality forces a Crabb–Choi–Crouzeix block. This identifies a concrete extremal mechanism behind the sharp constant $W(A)^\circ$4 (Hnetynkova et al., 2018).

4. Reductions, inheritance principles, and operator models

Several recent results reorganize the conjecture by showing that many matrix constructions preserve validity of the $W(A)^\circ$5-bound. If $W(A)^\circ$6 satisfies the conjecture, then so do all affine images $W(A)^\circ$7, the transpose $W(A)^\circ$8, and the adjoint $W(A)^\circ$9. The class is also closed under direct sums, and every rank-one matrix satisfies the conjecture. Tensoring with a normal matrix preserves the property as well: if $A$00 is normal and $A$01 satisfies the conjecture, then both $A$02 and $A$03 satisfy it (O'Loughlin et al., 2023).

These closure properties feed into a reduction to cyclic matrices. A vector $A$04 is cyclic for $A$05 if

$A$06

and $A$07 is cyclic if it has such a vector. The key reduction is that the full conjecture holds if and only if it holds for cyclic matrices. More precisely, if it holds for all cyclic matrices, then it holds for all matrices. In dimension $A$08, every non-cyclic matrix already satisfies the conjecture, because the extremal norm behavior compresses to a $A$09- or $A$10-dimensional subspace where the conjecture is known (O'Loughlin et al., 2023).

A functional-model reformulation makes the cyclic case more specific. If $A$11 is cyclic and $A$12 is cyclic for $A$13, then one defines

$A$14

with $A$15, and introduces the differentiation operator

$A$16

For cyclic $A$17, this operator is unitarily equivalent to $A$18. Consequently, Crouzeix’s conjecture holds in full generality if and only if it holds for these differentiation operators on finite-dimensional spaces of entire functions (O'Loughlin et al., 2023).

A further operator-model reduction appears for symmetric matrices. In dimension $A$19, the conjecture for symmetric matrices is equivalent to the conjecture for truncated Toeplitz operators on three-dimensional model spaces, and even to the analytic-symbol subclass. This ties one finite-dimensional corner of the conjecture directly to model-space operator theory (O'Loughlin et al., 2023).

5. Compressions of shifts, Blaschke products, and level-set variants

For a finite Blaschke product

$A$20

the model space

$A$21

has dimension $A$22, and the compressed shift

$A$23

provides a canonical finite-dimensional model operator. These compressions are central because they realize the completely non-unitary defect-one contractions and offer a highly structured testing ground for Crouzeix-type inequalities (Bickel et al., 2021).

A weaker but natural specialization is the level set Crouzeix conjecture (LSC). If $A$24 and $A$25 are finite Blaschke products with $A$26, then $A$27; if the full conjecture held for $A$28, it would force

$A$29

Equivalently, with

$A$30

the level-set formulation is

$A$31

A decisive sufficient condition is pseudohyperbolic-disk containment: if

$A$32

then LSC holds for every $A$33 of degree $A$34 (Bickel et al., 2021).

This framework has yielded several concrete results. LSC is proved for all unicritical test Blaschke products $A$35; for all pairs with unicritical $A$36 of degree $A$37 or $A$38; for every degree-$A$39 $A$40 against $A$41; and for the case where $A$42 and $A$43 has two connected components. The arguments rely on level-set rigidity for finite Blaschke products, pseudohyperbolic geometry, and explicit inner curves inside $A$44 (Bickel et al., 2021).

Subsequent work extends these methods to further families of model matrices and related nilpotent matrices. For the repeated-zero degree-$A$45 family

$A$46

the numerical range $A$47 contains a pseudohyperbolic disk of radius $A$48, which implies LSC for all $A$49. For the degree-$A$50 family

$A$51

one obtains a contained pseudohyperbolic disk of radius $A$52, again sufficient for LSC for all $A$53. There are also degree-$A$54 families

$A$55

for which, for each fixed $A$56, LSC holds for all $A$57 sufficiently close to $A$58, and explicit thresholds are given for $A$59 (Bickel et al., 2023).

The same circle of ideas also produces full, not merely level-set, Crouzeix results for some nilpotent matrices associated with repeated-zero model operators. For instance, certain $A$60 KMS-type nilpotent matrices satisfy the conjecture on explicit parameter intervals such as $A$61, and related families

$A$62

satisfy the conjecture for $A$63 on explicit $A$64-ranges obtained via similarity-to-Jordan-block estimates (Bickel et al., 2023).

6. Random matrices, algorithms, computation, and open directions

A recent asymptotic result verifies a restricted form of the conjecture for complex Ginibre matrices. If $A$65 is an $A$66 complex Ginibre matrix normalized so that the circular law fills the unit disk, then

$A$67

Equivalently, for every $A$68,

$A$69

eventually almost surely. This is not a proof of the full conjecture: it is restricted to a random ensemble, asymptotic in $A$70, probabilistic, and relies on the fact that $A$71 becomes nearly circular. The proof combines asymptotic circularity

$A$72

uniform resolvent bounds on annuli, and a perturbative disk argument in the Crouzeix–Palencia/Caldwell–Greenbaum–Li framework (Chen et al., 2023).

The same work shows how this asymptotic $A$73-spectral-set behavior enters GMRES. For systems

$A$74

the ideal-GMRES residual satisfies, almost surely as $A$75,

$A$76

with an alternative pseudospectral bound

$A$77

Here the numerical range yields the asymptotically optimal constant $A$78, while pseudospectra provide explicit spectral sets with usually worse constants (Chen et al., 2023).

Crouzeix-type bounds also appear in optimization. In nonlinear acceleration of multistep methods, including momentum and primal-dual schemes, the local linearized iteration operator $A$79 is generally non-symmetric. The RNA analysis reduces performance to norms of matrix polynomials $A$80, and Crouzeix’s inequality converts this into the planar approximation problem

$A$81

Consequently, acceleration performance is controlled by a Chebyshev problem on the numerical range of a nonsymmetric operator. This is the mechanism that extends numerical-range-based acceleration theory beyond symmetric fixed-point maps (Bollapragada et al., 2018).

Computational work on the Crouzeix ratio

$A$82

provides another perspective. Extensive nonsmooth optimization experiments, totaling nearly half a million runs and about $A$83 million evaluations, found many locally minimal values strictly between $A$84 and $A$85, but none below $A$86. The same locally minimal values frequently appear in both real and complex formulations, and approximate nonsmooth stationarity was verified numerically using near-active local maximizers on $A$87. These computations strongly support the conjecture that the global minimum of the Crouzeix ratio is $A$88, equivalently that the optimal universal constant is $A$89 (Overton, 2021).

A broader extension replaces the classical numerical range by the scaled $A$90-numerical range

$A$91

and develops spectral-set bounds through a generalized Crouzeix–Palencia framework. In this setting one proposed $A$92-analogue is

$A$93

which recovers the classical conjecture at $A$94 (O'Loughlin et al., 16 Mar 2026).

The central open problem remains the original universal $A$95-bound for arbitrary matrices. Several subsidiary questions sharpen the landscape. One asks whether the class of matrices satisfying the conjecture is closed under addition; a positive answer would imply the full conjecture, since rank-one matrices already satisfy it. Another asks whether extremal behavior can always be compressed to dimension $A$96, a mechanism that would force the conjecture by reduction to the $A$97 case. In the model-operator direction, open problems include characterizing extremal Blaschke-product degrees and the relation between norm-extremal and numerical-radius-extremal functions. In the random-matrix direction, it remains unknown whether large Ginibre matrices might asymptotically satisfy a better constant $A$98; numerics in that setting suggest the possibility of $A$99, but no proof is known (O'Loughlin et al., 2023).